app.py

import streamlit as st
import torch
import pickle
import numpy as np
from pathlib import Path
from sklearn.preprocessing import LabelEncoder
from torch.nn.functional import softmax
from model_folder.modeling_bilstm import BiLSTM
from tensorflow.keras.preprocessing.sequence import pad_sequences
import json
import pandas as pd

# Constants
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load Model and Tokenizer
@st.cache_resource
def load_model_and_tokenizer():
    model_path = Path("model_folder")
    tokenizer_path = model_path / "tokenizer.pkl"
    config_path = model_path / "config.json"
    model_weights_path = model_path / "pytorch_model.bin"

    # Load tokenizer
    with open(tokenizer_path, "rb") as f:
        tokenizer = pickle.load(f)

    # Load model config
    with open(config_path, "r") as f:
        config = json.load(f)

    # Get vocabulary size from the tokenizer
    vocab_size = len(tokenizer.word_index) + 1  # +1 for padding token

    # Initialize embedding matrix (vocab_size x embed_size)
    embedding_matrix = torch.zeros((vocab_size, config["embedding_dim"]))

    # Load model
    model = BiLSTM(
        embedding_matrix=embedding_matrix,
        embed_size=config["embedding_dim"],
        num_labels=config["num_labels"],
        hidden_size=config["hidden_size"],
        dropout=config["dropout_rate"]
    )
    model.load_state_dict(torch.load(model_weights_path, map_location=DEVICE))
    model.eval()
    model.to(DEVICE)

    # Label encoder with provided class names
    le = LabelEncoder()
    le.classes_ = np.array([
        "ADHD", "Acne", "Anxiety", "Bipolar Disorde", "Birth Control",
        "Depression", "Diabetes, Type 2", "Emergency Contraception",
        "High Blood Pressure", "Insomnia", "Obesity", "Pain",
        "Vaginal Yeast Infection", "Weight Loss"
    ])

    return model, tokenizer, le, config

# Predict Condition
def predict_condition(model, tokenizer, le, text, max_length):
    sequence = tokenizer.texts_to_sequences([text])
    padded_sequence = pad_sequences(sequence, maxlen=max_length)
    tensor = torch.tensor(padded_sequence, dtype=torch.long).to(DEVICE)
    
    with torch.no_grad():
        logits = model(tensor)
        probabilities = softmax(logits, dim=1)
        predicted_class = torch.argmax(probabilities, axis=1).item()
    
    predicted_condition = le.inverse_transform([predicted_class])[0]
    confidence = probabilities[0][predicted_class].item()
    
    return predicted_condition, confidence

# Recommend Drugs
def recommend_drugs(predicted_condition, df, top_n=5):
    # Filter the dataset for the predicted condition
    condition_df = df[df['condition'] == predicted_condition]

    # Filter for positive reviews (e.g., rating >= 8)
    positive_reviews_df = condition_df[condition_df['rating'] >= 8]

    # Group by drug and calculate the number of positive reviews and average rating
    drug_stats = positive_reviews_df.groupby('drugName').agg(
        num_reviews=('rating', 'size'),  # Count the number of reviews
        avg_rating=('rating', 'mean')    # Calculate the average rating
    ).reset_index()

    # Sort by number of reviews and average rating
    drug_stats = drug_stats.sort_values(by=['num_reviews', 'avg_rating'], ascending=False)

    # Return the top N drugs
    return drug_stats.head(top_n)

# Streamlit UI
def main():
    st.title("Condition Prediction and Drug Recommendation")
    st.write("Enter symptoms below to predict the medical condition and get drug recommendations.")

    # Load the dataset
    df = pd.read_csv('dataset/custom.csv')

    model, tokenizer, le, config = load_model_and_tokenizer()

    user_input = st.text_area("Enter your symptoms", "")
    if st.button("Predict"):
        if user_input.strip():
            with st.spinner("Predicting..."):
                predicted_condition, confidence = predict_condition(model, tokenizer, le, user_input, config["max_length"])
                st.success(f"Predicted Condition: {predicted_condition} (Confidence: {confidence:.2f})")

                # Get drug recommendations
                top_drugs = recommend_drugs(predicted_condition, df, top_n=5)
                
                # Display recommended drugs
                st.write("**Top Recommended Drugs**")
                if not top_drugs.empty:
                    for i, row in top_drugs.iterrows():
                        st.write(f"- {row['drugName']} (Positive Reviews: {row['num_reviews']}, Avg Rating: {row['avg_rating']:.2f})")
                else:
                    st.write("No drugs found for this condition.")

        else:
            st.warning("Please enter a review.")

if __name__ == "__main__":
    main()